Foreshocks are known to occur before certain large earthquakes, but the physical mechanism is still in debate. Recent field and laboratory studies have supported either cascade‐triggering, pre‐slip or a combination of both models. Here we use a dense seismic and geodetic network deployed in 2018 in Southwest China to quantify the long‐term background seismicity, short‐term spatio‐temporal evolutions of foreshocks and transient deformation before the 2021 MW6.1 Yangbi earthquake. We find multiple episodes of migrating foreshocks and repeating earthquakes in the last 4 days of the Yangbi mainshock. A rapid migration of microseismicity started in the last 30 minutes toward the eventual initiation point of the Yangbi mainshock, well beyond the rupture zone of the largest MW5.2 foreshock at that time. The Coulomb stress changes due to relatively large‐size foreshocks at the mainshock hypocenter are positive but relatively small (0.005–0.05 MPa). We also observe continuous geodetic signals with ∼20 mm of cumulative displacement at a nearby GPS station coinciding with the last hour of intense foreshocks. The inverted aseismic moment is equivalent to a MW5.56 event with the maximum slip of ∼200 mm at 6 km depth, larger than the cumulative seismic slip of ∼60–120 mm inverted from waveforms of several large M4+ foreshocks. In addition, the predicted surface displacements based on the coseismic slips of the M4+ foreshocks are unobservable in the GPS data. We propose a migratory slow‐slip model where a transient slow‐slip event drives the last hour of foreshock sequence and eventually trigger the Yangbi mainshock. A dense seismic network and geodetic network around the 2021 MW6.1 Yangbi earthquake in Southwest China captured a multiple stage of its foreshock within 4 days along its rupture fault. More than 1,800 foreshocks occurred in the first two stages, and the seismicity shows systematic expansion along the fault strike following large foreshocks in both stages. The last stage contained 130 events within the last hour before the Yangbi mainshock, and it started with a MW4.3 event, followed by the largest foreshock with MW= 5.2. We calculate stress changes due to relatively large‐size foreshocks, and find that we cannot use the cascade‐triggering model to explain how the mainshock was triggered by those foreshocks. Continuous geodetic signals with ∼20 mm cumulative displacement are observable at the nearest GPS station starting in the last hour before the mainshock. The inverted aseismic moment is equivalent to a MW5.56 event with maximum slip of ∼200 mm, larger than the cumulative seismic slip of several large M4+ foreshocks. In contrast, the predicted surface displacements based on the coseismic slips of the M4+ foreshocks are negligible at the GPS stations. Our results suggest that migratory foreshocks driven by aseismic slip are observable on a continental strike‐slip earthquake. Multiple episodes of migrating foreshocks and repeating events are found right before the 2021 MW6.1 Yangbi mainshockHigh‐rate GPS time series revealed a MW5.56 slow‐slip event in the last 1 hr before the Yangbi mainshockWe propose a migratory slow‐slip model to explain the Yangbi foreshock sequence Multiple episodes of migrating foreshocks and repeating events are found right before the 2021 MW6.1 Yangbi mainshock High‐rate GPS time series revealed a MW5.56 slow‐slip event in the last 1 hr before the Yangbi mainshock We propose a migratory slow‐slip model to explain the Yangbi foreshock sequence